hello

[illumos-gate] 16096 want proc_get_lwpstatus(3PROC) -- Robert Mustacchi <rm⊙fo>

[illumos-gate] 16107 asy VERIFY too strong in face of polledio -- Patrick Mooney <pmooney⊙pc>

Happy new CPU bug! vusec.net/projects/slam (though it only affects Intel & ARM CPUs that haven't shipped to the public yet, but some AMD ones that have)

my complete slacking on ARM paging pays off again!

Storing information in pointers... sheesh, Macintosh System 1 did that, for crying out loud!

heh, I remember the 24-bit -> 32-bit pointer change in the old Mac Systems back when I had a Mac II

of course, we do that on recent SPARC systems for pointer checking with ADI, but that's more like ARM's Memory Tagging than ARM's Top-Byte Ignore

(I haven't looked at memory tagging either!)

I looked at pointer auth tho

didn't do anything, but I looked

danmcd: I mean, that's why OI force an /etc/system tweak to leave some bits free for javascript runtimes etc.

TBI is basically doing that, but without having to mask them off constantly

(at the expense of what alan's link talks about, re: address canonicity.)

(to be clear, the /etc/system tweak is so we don't put runtimes at an address that _requires a bit they use for type tagging_)

@alanc gets it. :) And thank you richlowe for the clarification.

and I remember the mac brouhaha, but thankfully only second hand

and I was impressed with amd64 making the VA hole checked and very cautious

so it's impossible to try to steal those bits, and get screwed the same way

right, we have to do that with javascript runtimes as well, many of which assume only 48 bits of pointers matter, and won't be happy with 53-bit address busses

though we do it via linker flags instead of /etc/system

actual flags, or something like map.above4G etc?

(well, linker flags to load a special mapfile)

that makes more sense

maybe shipping that in /usr/lib/ld would make sense?

and dare I ask about github.com/oracle/solaris-userland/…ts/gnome/gnome-shell/mapfile.js#L3? :)

we don't want to encourage it, but I suppose it's too late to pretend no one should ever need it

my logic was "It's probably better to ship one, that does it properly, then keep adding them to solaris-userland and also not helping 3rd parties"

than*

yeah

but it's not like I did that for us :)

what about that line did you want to ask?

it sets the p_vaddr in the Program Header to the start of the range that overlaps with the bits set in the pointers

Program Header[5]: spidermonkey_reserve

p_vaddr: 0x800000000000 p_flags: 0

p_paddr: 0 p_type: [ PT_SUNW_RESERVE ]

p_filesz: 0 p_memsz: 0xffff7fffffff0000

p_offset: 0 p_align: 0x100000

alanc: sorry, the line-link was an accent. I meant `mapfile.js`

accident, even

gnome-shell uses libmozjs to run javascript code - that's what gnome-shell extensions are written in

ah, so it's a mapfile for javascript

so same javascript engine as firefox, same problem with addresses

named so as to piss off editor syntax selection

ah, yes, hehe

not a new syntax for mapfiles, just following the convention of mapfile.<whats-it-for>

though I guess /usr/lib/ld just uses map.<whats-it-for> as a naming convention instead